WM 2.2 not responded in tiled import

Hi!
WM 2.2 not responded in tiled import. We have 36 RGB tiles in 512*512 resolution in png format.
In 20 minutes we see WM memory increasing to 1.5 GB and no result.
Our system - win7, 4 GB RAM. We have same situation on 2 another PC`s
How to solve this problem?

Something is certainly not right here. Can you walk me through your work flow to reproduce this issue?

Typical steps:

  1. Drop a tiled file input device into the world
  2. Choose “specify from files…” to find your tileset, and select any two tiles in the tileset
  3. Once you click OK in step 2, WM will preload your tileset into memory, and you should see statistics about the tileset appear. Do you reach this point?

If you do, can you please tell me what it reads under the Tile Extents, Format, and Resolution readouts, and also under the Cache Performance Counters readouts?

We use these typical steps.
After step 2 our fields are:
Tile Extents - 6 by 6 (36 total)
Format - PNG
Resolution readouts - 5000x5000 (30000x30000)
Cache Performance Counters
Hits - 0
Misses - 0
Evictions - 0
Protected Levels - (ALL)

Its strange for me, because now WM could not load in tiled node even low-res pieces 312x312px, after pressing Ok button the program simply frezes.

Hello again,

So, my first question is: Are the tiles 5000x5000 each or 512x512 each? Based on the readout you’ve posted, WM believes they are 5k x 5k.

Now, if your tiles really are 512x512 and not 5k x 5k as WM thinks, then there is definitely a bug here where the PNG loader is not representing things correctly! I have not been able to reproduce any such bug, but there are two things to try in this case.

  1. If you can, batch-convert the PNGs to another supported import format such as TIFF and see if it still has the same misunderstanding
  2. You can send a single tile of the input PNG to me at support@world-machine.com to examine and discover why WM is misinterpreting.

On the other hand, if your tileset is actually 36 5k x 5k tiles, then you are probably seeing normal behavior. A 36 tile 5k x 5k array takes up large amounts of memory (3GB for heightfield, 12GB for RGB), and the TFI device must load and establish a cached working set of those tiles at some down sampled resolution before it can begin.

The precaching is a one-shot process when the world file is loaded (or the TFI device created), and should not affect your use of WM once its done, but can take some time ( a couple minutes ) if you have a very large tileset.

One thing that can immediately help is to go to Performance Cache and slide the memory allocated to it as far to the right as you can. This will reduce the cache evictions during precaching and increase performance during builds as more of the tileset will be held in memory at once.

I’ve tested the TFI with far larger datasets even (approaching 100,000 by 100,000) and other than a long pre-cache time, you should have no issues concerning size of the tileset.

:stuck_out_tongue: Today i figured out why this problem was happening in World Machine, it was not the case of huge resolutions, instead it seems that WM does not like the overlapping maps which i`ve exported from Global Mapper, also it seems it does not likes textures cropped in Global Mapper too, its best when pieces have square proportions, even if they handle only one border pixel of the map.

That’s a good point – all my test cases so far are with square tiles. I will add rectangular areas to the test set and make sure things are OK with that on the next release.

WM tiling is based upon all tiles having identical resolutions; if this is not the case it will definitely cause issues. I will honestly admit I don’t know how well that error path is being handled right now, so I will check that as well.

Cheers

G-Rom, something I’ve found with Global Mapper is that if you export to a bounding box, you can sometimes get null values in the tiles on the edges- or very out of range values. It’s caused by the reprojection of pixel samples.

monks

Thanks for answers, probably i`ll create a new thread where we can discuss problems with GM->World Machine Pipeline because some things which are happening seems not so smooth and clear.